Case study
Performance Optimization After Elasticsearch Migration for Matchmaking Platforms
How SquareShift resolved post-migration performance issues for a leading matchmaking platform in Asia, more than doubling Elasticsearch query performance under 90,000 requests per second.
A leading matchmaking platform in Asia serving more than 35 million users through web and mobile apps.
Real-time search and massive user engagement mean the platform depends on Elasticsearch for its core matchmaking intelligence.
Users served through the platform's real-time matchmaking search. Improvement in Elasticsearch query latency after tuning. Requests per second sustained under the retuned cluster.
Key servicesHow SquareShift delivered it.
The challenge
The client, a leading matchmaking platform in Asia serving more than 35 million users, had already migrated its Elasticsearch cluster from AWS to GCP when performance started slipping. Query latency degraded and the 47-node cluster wasn’t delivering the throughput the platform’s real-time, high-engagement search needed.
The client’s in-house team didn’t have the depth of Elastic tuning experience to diagnose whether the problem was hardware sizing, cluster layout, or something else in the new environment.
What we delivered
SquareShift compared the AWS and GCP configurations side by side to isolate the resource mismatch driving the slowdown, then ran cluster analysis and benchmarking with ES Rally to test fixes under real load.
The team recommended and implemented optimal node configurations for the GCP environment, then improved cluster layout and indexing patterns across all 47 nodes.
The payoff
Query latency improved by more than 2x at a sustained load of 90,000 requests per second, giving the platform’s real-time matchmaking search the throughput its 35 million users need. The retuned cluster now matches its configuration to how GCP actually performs, not how the old AWS setup did.
A cloud migration that skips re-benchmarking hardware sizing invites a latency surprise — the fix isn't more nodes, it's matching node layout to how the new cloud performs.
Cloud Modernization Practice Lead, SquareShift
Where this work sits